The Students Walking Away From Computer Science Are Telling Us Something

For twenty years, computer science was the safe answer for a smart eighteen-year-old who wanted a well-paid career: learn to code, land a junior job, build experience, move up.

That answer now comes with a large asterisk.

In a February 2026 feature, The Guardian spoke to students switching from computer science into nursing and paramedicine, alongside white-collar workers considering skilled trades. It also cited a survey where 53% of young respondents were considering blue-collar or skilled-trade careers to reduce their exposure to AI.

That figure does not mean half of a generation has abandoned software. A survey about consideration is not an enrolment count, and fear is not a labour-market forecast. But the signal deserves more attention than the standard reply from tech leaders: AI will make everyone more productive, so keep learning to code.

Students are looking past the productivity pitch and asking whether the first rung of the ladder will still exist when they graduate.

A task can disappear while the job title survives

The Guardian also cited ADP payroll data from December 2025 showing employment fell in professional and business services and in IT, while healthcare and hospitality grew. One month of payroll data cannot prove AI caused those shifts. Hiring cycles, interest rates, sector demand and ordinary corporate cost-cutting all matter.

The figures do match a more specific concern. AI does not need to erase an occupation to damage its entry route. It only needs to absorb enough routine work that employers hire fewer beginners.

Software is the clearest example. A junior developer once earned trust by writing small features, fixing obvious bugs, producing tests and learning a codebase through repetition. Models can now perform much of that work quickly. An experienced engineer still has to understand the system, inspect output, weigh trade-offs and own the consequences. But if a model handles the practice work, where does the next experienced engineer come from?

A thoughtful April essay from Daydreams in Ruby describes the same bargain. Developers can hand mechanical tasks to AI and spend more time on architecture, security, review and integration. The risk is skill atrophy: engineers stop practising the work that once built the judgment needed for those higher-level decisions.

That is the pressure students see. The senior role remains visible; the route into it is getting harder to read.

The same pattern reaches paralegals, analysts, accountants and junior managers. A company can retain the job title while automating research, first drafts, reconciliations and routine reporting. The remaining work carries more context and more responsibility. Employers then demand senior judgment while shrinking the pool of junior tasks that used to produce it.

This is a training problem disguised as a productivity win.

Human value is moving toward consequence

A 2025 paper, Future of Work with AI Agents, examined 844 tasks across 104 occupations. Its useful move was to compare technical capability with worker preference: what can an AI agent do, and what do workers want it to do? The authors found a shift in the competencies that matter, away from information handling and toward interpersonal skills.

That result is easy to soften into vague advice about communication. The harder point is accountability. Information becomes cheaper when a model can produce a plausible draft in seconds. Decisions remain expensive because someone must understand the context, defend the choice and carry the cost when it fails.

Sam Altman made the distinction unusually clear. In July 2026, Business Insider reported his view that people still want a human CEO. People want a person accountable for important decisions, he said, and often prefer dealing with other people. This came after his earlier suggestion that OpenAI might one day have an AI chief executive.

The retreat matters less as a prediction about executive suites than as a clue about the labour market. The safest work is not necessarily the most intelligent or prestigious. It is work bound tightly to consequence: a named person makes the call, deals with the people affected and cannot vanish behind an automated answer.

Nursing has that shape. A nurse works with a patient in a room, notices changes that may not fit a form, acts under clinical rules and answers to colleagues, patients and regulators. Electrical work also binds diagnosis to a physical site, trained hands, safety rules and licensing. AI will change both fields, but their tasks are harder to detach from the worker and run elsewhere.

Calling these careers “AI-proof” would be careless. Better diagnostic systems, robotics and remote operations will alter physical and care work too. The difference today is task separability. Many desk jobs produce digital artifacts that can be generated, checked and moved without the worker being present. Bedside care and fault-finding in a live building resist that separation.

Computer science needs a better first rung

Software still has enormous unsolved demand, and cheaper code makes more systems economical to build. The opportunity is real, but the old training contract is under strain.

Computer science courses should respond by teaching students to work on systems where model output meets consequence. That means reading unfamiliar code, designing tests that catch plausible mistakes, tracing failures across services, modelling security threats, checking data provenance and explaining a technical decision to someone who will rely on it. Students still need to write code themselves, because verification without underlying skill quickly becomes approval by vibes.

Employers have a harder job. If AI removes a chunk of junior production work, they must create deliberate apprenticeships rather than assume senior engineers will appear from nowhere. Give new hires bounded systems to own. Require them to inspect model output, diagnose failures and defend changes in review. Measure whether they can find the wrong answer, not only whether they can ship the generated one.

The students in The Guardian are not reliable prophets of which occupations will win, but they are reliable customers of the career paths on offer. Right now, some see a clearer route from novice to trusted professional in nursing or a trade than in software.

Computer science can earn them back if it rebuilds a credible first rung and teaches the judgment that senior roles demand.